Several fixes for the Sun DBRI driver
PATCH 1/1 Several fixes for the Sun DBRI driver
This patch contains the following fixes to the Alsa DBRI driver:
- Remove 2.6.13 build warning on the prom_getproperty() call.
- Rework command synchronization: send a sequence number with D_WAIT, and check it in the completion interrupt.
Move synchronization delays from _cmdsend() to _cmdlock() allowing the CPU to do other usefull things while the DBRI is processing the commands.
- Fix first argument of printk() calls.
- Enable burst transfers for DBRI. Original 2.4 patch from Krzysztof Helt
- Make dbri_debug module parameter writable from sysfs. Remove obsolete write access to the /proc debug file.
- Replace udelay() with msleep_interruptible() where possible.
- Update documentation comments.
This diff is against 2.6.13, i.e. ALSA 1.0.9b. I can also send each fix split out if you want. Please apply.
Thanks,
Martin
---
sound/sparc/dbri.c renamed from sound/sparc/dbri.c.orig +103 -102
@@ -1,6 +1,6 @@/* * Driver for DBRI sound chip found on Sparcs. -* Copyright (C) 2004 Martin Habets ([email protected]) +* Copyright (C) 2004, 2005 Martin Habets ([email protected]) * * Based entirely upon drivers/sbus/audio/dbri.c which is: * Copyright (C) 1997 Rudolf Koenig ([email protected])
@@ -43,6 +43,12 @@* audio devices. But the SUN HW group decided against it, at least on my * LX the speakerbox connector has at least 1 pin missing and 1 wrongly * connected. +* +* I've tried to stick to the following function naming conventions: +* snd_* ALSA stuff +* cs4215_* CS4215 codec specfic stuff +* dbri_* DBRI high-level stuff +* other DBRI low-level stuff */ #include <sound/driver.h>
@@ -87,7 +93,7 @@#define D_DESC (1<<5)static int dbri_debug =0 ; -module_param (dbri_debug, int,0444 ); +module_param (dbri_debug, int,0644 );MODULE_PARM_DESC (dbri_debug,"Debug value for Sun DBRI soundcard." );#ifdef DBRI_DEBUG
@@ -320,7 +326,8 @@void __iomem *regs;/* dbri HW regs */ int dbri_version;/* 'e' and up is OK */ int dbri_irqp;/* intr queue pointer */ -int wait_seen; +int wait_send;/* sequence of command buffers send */ +int wait_ackd;/* sequence of command buffers acknowledged */ struct dbri_pipe pipes[DBRI_NO_PIPES ];/* DBRI's 32 data pipes */ struct dbri_desc descs[DBRI_NO_DESCS ];
@@ -625,16 +632,13 @@ Commands are sent to the DBRI by building a list of them in memory, then writing the address of the first list item to DBRI register 8. -The list is terminated with a WAIT command, which can generate a -CPU interrupt if required. +The list is terminated with a WAIT command, which generates a +CPU interrupt to signal completion. Since the DBRI can run in parallel with the CPU, several means of -synchronization present themselves. The original scheme (Rudolf's) -was to set a flag when we "cmdlock"ed the DBRI, clear the flag when -an interrupt signaled completion, and wait on a wait_queue if a routine -attempted to cmdlock while the flag was set. The problems arose when -we tried to cmdlock from inside an interrupt handler, which might -cause scheduling in an interrupt (if we waited), etc, etc +synchronization present themselves. The method implemented here is close +to the original scheme (Rudolf's), and uses 2 counters (wait_send and +wait_ackd) to synchronize the command buffer between the CPU and the DBRI. A more sophisticated scheme might involve a circular command buffer or an array of command buffers. A routine could fill one with
@@ -642,70 +646,75 @@completion of the current command buffer, look on the list for the next one. -I've decided to implement something much simpler - after each command, -the CPU waits for the DBRI to finish the command by polling the P bit -in DBRI register 0. I've tried to implement this in such a way -that might make implementing a more sophisticated scheme easier. -Every time a routine wants to write commands to the DBRI, it must first call dbri_cmdlock() and get an initial pointer into dbri->dma->cmd -in return. After the commands have been writen, dbri_cmdsend() is -called with the final pointer value. +in return. dbri_cmdlock() will block if the previous commands have not +been completed yet. After this the commands can be written to the buffer, +and dbri_cmdsend() is called with the final pointer value to send them +to the DBRI. */ +static void dbri_process_interrupt_buffer (snd_dbri_t * dbri); +enum dbri_lock_t { NoGetLock, GetLock }; +#define MAXLOOPS 10static volatile s32 *dbri_cmdlock (snd_dbri_t * dbri,enum dbri_lock_t get) { +int maxloops =MAXLOOPS ; +#ifndef SMP if ((get == GetLock) &&spin_is_locked (&dbri->lock)) {printk (KERN_ERR "DBRI: cmdlock called while in spinlock." ); }#endif +/* Delay if previous commands are still being processed */ +while ((--maxloops) >0 && (dbri->wait_send != dbri->wait_ackd)) { +msleep_interruptible (1 ); +/* If dbri_cmdlock() got called from inside the +* interrupt handler, this will do the processing. +*/ +dbri_process_interrupt_buffer (dbri); + } +if (maxloops ==0 ) { +printk (KERN_ERR "DBRI: Chip never completed command buffer %d\n" , + dbri->wait_send); + }else { +dprintk (D_CMD ,"Chip completed command buffer (%d)\n" , +MAXLOOPS - maxloops -1 ); + } +/*if (get == GetLock) spin_lock(&dbri->lock); */ return &dbri->dma->cmd[0 ]; } -static void dbri_process_interrupt_buffer (snd_dbri_t *); -static void dbri_cmdsend (snd_dbri_t * dbri,volatile s32 * cmd) { -int MAXLOOPS =1000000 ; -int maxloops =MAXLOOPS ;volatile s32 *ptr; +u32 reg;for (ptr = &dbri->dma->cmd[0 ]; ptr < cmd; ptr++) {dprintk (D_CMD ,"cmd: %lx:%08x\n" , (unsigned long )ptr, *ptr); }if ((cmd - &dbri->dma->cmd[0 ]) >=DBRI_NO_CMDS -1 ) { -printk ("DBRI: Command buffer overflow! (bug in driver)\n" ); +printk (KERN_ERR "DBRI: Command buffer overflow! (bug in driver)\n" );/* Ignore the last part. */ cmd = &dbri->dma->cmd[DBRI_NO_CMDS -3 ]; } + dbri->wait_send++; + dbri->wait_send &=0xffff ;/* restrict it to a 16 bit counter. */ *(cmd++) =DBRI_CMD (D_PAUSE ,0 ,0 ); - *(cmd++) =DBRI_CMD (D_WAIT ,1 ,0 ); - dbri->wait_seen =0 ; + *(cmd++) =DBRI_CMD (D_WAIT ,1 , dbri->wait_send); + +/* Set command pointer and signal it is valid. */ sbus_writel (dbri->dma_dvma, dbri->regs +REG8 ); -while ((--maxloops) >0 && (sbus_readl (dbri->regs +REG0 ) &D_P )) -barrier (); -if (maxloops ==0 ) { -printk (KERN_ERR "DBRI: Chip never completed command buffer\n" ); -dprintk (D_CMD ,"DBRI: Chip never completed command buffer\n" ); - }else { -while ((--maxloops) >0 && (!dbri->wait_seen)) -dbri_process_interrupt_buffer (dbri); -if (maxloops ==0 ) { -printk (KERN_ERR "DBRI: Chip never acked WAIT\n" ); -dprintk (D_CMD ,"DBRI: Chip never acked WAIT\n" ); - }else { -dprintk (D_CMD ,"Chip completed command " -"buffer (%d)\n" ,MAXLOOPS - maxloops); - } - } + reg =sbus_readl (dbri->regs +REG0 ); + reg |=D_P ; +sbus_writel (reg, dbri->regs +REG0 );/*spin_unlock(&dbri->lock); */ }
@@ -757,10 +766,11 @@for (n =0 ; n <DBRI_NO_PIPES ; n++) dbri->pipes[n].desc = dbri->pipes[n].first_desc =-1 ; -/* We should query the openprom to see what burst sizes this -* SBus supports. For now, just disable all SBus bursts */ +/* A brute approach - DBRI falls back to working burst size by itself +* On SS20 D_S does not work, so do not try so high. */ tmp =sbus_readl (dbri->regs +REG0 ); - tmp &= ~(D_G |D_S |D_E ); + tmp |=D_G |D_E ; + tmp &= ~D_S ;sbus_writel (tmp, dbri->regs +REG0 );/*
@@ -805,13 +815,13 @@volatile int *cmd;if (pipe <0 || pipe >31 ) { -printk ("DBRI: reset_pipe called with illegal pipe number\n" ); +printk (KERN_ERR "DBRI: reset_pipe called with illegal pipe number\n" );return ; } sdp = dbri->pipes[pipe].sdp;if (sdp ==0 ) { -printk ("DBRI: reset_pipe called on uninitialized pipe\n" ); +printk (KERN_ERR "DBRI: reset_pipe called on uninitialized pipe\n" );return ; }
@@ -834,12 +844,12 @@static void setup_pipe (snd_dbri_t * dbri,int pipe,int sdp) {if (pipe <0 || pipe >31 ) { -printk ("DBRI: setup_pipe called with illegal pipe number\n" ); +printk (KERN_ERR "DBRI: setup_pipe called with illegal pipe number\n" );return ; }if ((sdp &0xf800 ) != sdp) { -printk ("DBRI: setup_pipe called with strange SDP value\n" ); +printk (KERN_ERR "DBRI: setup_pipe called with strange SDP value\n" );/* sdp &= 0xf800; */ }
@@ -872,13 +882,13 @@int nextpipe;if (pipe <0 || pipe >31 || basepipe <0 || basepipe >31 ) { -printk - ("DBRI: link_time_slot called with illegal pipe number\n" ); +printk (KERN_ERR +"DBRI: link_time_slot called with illegal pipe number\n" );return ; }if (dbri->pipes[pipe].sdp ==0 || dbri->pipes[basepipe].sdp ==0 ) { -printk ("DBRI: link_time_slot called on uninitialized pipe\n" ); +printk (KERN_ERR "DBRI: link_time_slot called on uninitialized pipe\n" );return ; }
@@ -960,8 +970,8 @@int val;if (pipe <0 || pipe >31 || prevpipe <0 || prevpipe >31 ) { -printk - ("DBRI: unlink_time_slot called with illegal pipe number\n" ); +printk (KERN_ERR +"DBRI: unlink_time_slot called with illegal pipe number\n" );return ; }
@@ -1001,22 +1011,22 @@volatile s32 *cmd;if (pipe <16 || pipe >31 ) { -printk ("DBRI: xmit_fixed: Illegal pipe number\n" ); +printk (KERN_ERR "DBRI: xmit_fixed: Illegal pipe number\n" );return ; }if (D_SDP_MODE (dbri->pipes[pipe].sdp) ==0 ) { -printk ("DBRI: xmit_fixed: Uninitialized pipe %d\n" , pipe); +printk (KERN_ERR "DBRI: xmit_fixed: Uninitialized pipe %d\n" , pipe);return ; }if (D_SDP_MODE (dbri->pipes[pipe].sdp) !=D_SDP_FIXED ) { -printk ("DBRI: xmit_fixed: Non-fixed pipe %d\n" , pipe); +printk (KERN_ERR "DBRI: xmit_fixed: Non-fixed pipe %d\n" , pipe);return ; }if (!(dbri->pipes[pipe].sdp &D_SDP_TO_SER )) { -printk ("DBRI: xmit_fixed: Called on receive pipe %d\n" , pipe); +printk (KERN_ERR "DBRI: xmit_fixed: Called on receive pipe %d\n" , pipe);return ; }
@@ -1036,17 +1046,17 @@static void recv_fixed (snd_dbri_t * dbri,int pipe,volatile __u32 * ptr) {if (pipe <16 || pipe >31 ) { -printk ("DBRI: recv_fixed called with illegal pipe number\n" ); +printk (KERN_ERR "DBRI: recv_fixed called with illegal pipe number\n" );return ; }if (D_SDP_MODE (dbri->pipes[pipe].sdp) !=D_SDP_FIXED ) { -printk ("DBRI: recv_fixed called on non-fixed pipe %d\n" , pipe); +printk (KERN_ERR "DBRI: recv_fixed called on non-fixed pipe %d\n" , pipe);return ; }if (dbri->pipes[pipe].sdp &D_SDP_TO_SER ) { -printk ("DBRI: recv_fixed called on transmit pipe %d\n" , pipe); +printk (KERN_ERR "DBRI: recv_fixed called on transmit pipe %d\n" , pipe);return ; }
@@ -1075,12 +1085,12 @@int last_desc =-1 ;if (info->pipe <0 || info->pipe >15 ) { -printk ("DBRI: setup_descs: Illegal pipe number\n" ); +printk (KERN_ERR "DBRI: setup_descs: Illegal pipe number\n" );return -2 ; }if (dbri->pipes[info->pipe].sdp ==0 ) { -printk ("DBRI: setup_descs: Uninitialized pipe %d\n" , +printk (KERN_ERR "DBRI: setup_descs: Uninitialized pipe %d\n" , info->pipe);return -2 ; }
@@ -1090,20 +1100,20 @@if (streamno ==DBRI_PLAY ) {if (!(dbri->pipes[info->pipe].sdp &D_SDP_TO_SER )) { -printk ("DBRI: setup_descs: Called on receive pipe %d\n" , +printk (KERN_ERR "DBRI: setup_descs: Called on receive pipe %d\n" , info->pipe);return -2 ; } }else {if (dbri->pipes[info->pipe].sdp &D_SDP_TO_SER ) { -printk - ("DBRI: setup_descs: Called on transmit pipe %d\n" , +printk (KERN_ERR +"DBRI: setup_descs: Called on transmit pipe %d\n" , info->pipe);return -2 ; }/* Should be able to queue multiple buffers to receive on a pipe */ if (pipe_active (dbri, info->pipe)) { -printk ("DBRI: recv_on_pipe: Called on active pipe %d\n" , +printk (KERN_ERR "DBRI: recv_on_pipe: Called on active pipe %d\n" , info->pipe);return -2 ; }
@@ -1120,7 +1130,7 @@break ; }if (desc ==DBRI_NO_DESCS ) { -printk ("DBRI: setup_descs: No descriptors\n" ); +printk (KERN_ERR "DBRI: setup_descs: No descriptors\n" );return -1 ; }
@@ -1165,7 +1175,7 @@ }if (first_desc ==-1 || last_desc ==-1 ) { -printk ("DBRI: setup_descs: Not enough descriptors available\n" ); +printk (KERN_ERR "DBRI: setup_descs: Not enough descriptors available\n" );return -1 ; }
@@ -1270,7 +1280,7 @@int divisor =12288 / clockrate;if (divisor >255 || divisor * clockrate !=12288 ) -printk ("DBRI: illegal bits_per_frame in setup_chi\n" ); +printk (KERN_ERR "DBRI: illegal bits_per_frame in setup_chi\n" ); *(cmd++) =DBRI_CMD (D_CHI ,0 ,D_CHI_CHICM (divisor) |D_CHI_FD |D_CHI_BPF (bits_per_frame));
@@ -1474,7 +1484,6 @@/* Temporarily mute outputs, and wait 1/8000 sec (125 us) * to make sure this takes. This avoids clicking noises. */ -cs4215_setdata (dbri,1 );udelay (125 );
@@ -1530,8 +1539,8 @@ tmp |=D_C ;/* Enable CHI */ sbus_writel (tmp, dbri->regs +REG0 ); -for (i =64 ; ((dbri->mm.status &0xe4 ) !=0x20 ); --i) { -udelay (125 ); +for (i =10 ; ((dbri->mm.status &0xe4 ) !=0x20 ); --i) { +msleep_interruptible (1 ); }if (i ==0 ) { dprintk(D_MM, "CS4215 didn't respond to CLB (0x%02x)\n",
@@ -1678,8 +1687,8 @@Complicated interrupts are handled by dedicated functions (which appear first in this file). Any pending interrupts can be serviced by calling dbri_process_interrupt_buffer(), which works even if the CPU's -interrupts are disabled. This function is used by dbri_cmdsend() -to make sure we're synced up with the chip after each command sequence, +interrupts are disabled. This function is used by dbri_cmdlock() +to make sure we're synced up with the chip before each command sequence, even if we're running cli'ed. */
@@ -1765,11 +1774,13 @@* Called by main interrupt handler when DBRI signals transmission complete * on a pipe (interrupt triggered by the B bit in a transmit descriptor). * -* Walks through the pipe's list of transmit buffer descriptors, releasing -* each one's DMA buffer (if present), flagging the descriptor available, -* and signaling its callback routine (if present), before proceeding -* to the next one. Stops when the first descriptor is found without +* Walks through the pipe's list of transmit buffer descriptors and marks +* them as available. Stops when the first descriptor is found without * TBC (Transmit Buffer Complete) set, or we've run through them all. +* +* The DMA buffers are not released, but re-used. Since the transmit buffer +* descriptors are not clobbered, they can be re-submitted as is. This is +* done by the xmit_descs() tasklet above since that could take longer. */ static void transmission_complete_intr (snd_dbri_t * dbri,int pipe)
@@ -1885,7 +1896,11 @@ }if (channel ==D_INTR_CMD && command ==D_WAIT ) { - dbri->wait_seen++; + dbri->wait_ackd = val; +if (dbri->wait_send != val) { +printk (KERN_ERR "Processing wait command %d when %d was send.\n" , + val, dbri->wait_send); + }return ; }
@@ -1994,8 +2009,7 @@* The only one I've seen is MRR, which will be triggered * if you let a transmit pipe underrun, then try to CDP it. * -* If these things persist, we should probably reset -* and re-init the chip. +* If these things persist, we reset the chip. */ if ((++errcnt) %10 ==0 ) {dprintk (D_INT ,"Interrupt errors exceeded.\n" );
@@ -2094,7 +2108,7 @@if ((ret =snd_pcm_lib_malloc_pages (substream,params_buffer_bytes (hw_params))) <0 ) { -snd_printk (KERN_ERR "malloc_pages failed with %d\n" , ret); +printk (KERN_ERR "malloc_pages failed with %d\n" , ret);return ret; }
@@ -2455,8 +2469,7 @@for (idx =0 ; idx <NUM_CS4215_CONTROLS ; idx++) {if ((err =snd_ctl_add (card, -snd_ctl_new1 (&dbri_controls[idx], - dbri))) <0 ) +snd_ctl_new1 (&dbri_controls[idx], dbri))) <0 )return err; }
@@ -2490,8 +2503,6 @@int pipe;snd_iprintf (buffer,"debug=%d\n" , dbri_debug); -snd_iprintf (buffer,"CHI pipe in=%d, out=%d\n" , - dbri->chi_in_pipe, dbri->chi_out_pipe);for (pipe =0 ; pipe <32 ; pipe++) {if (pipe_active (dbri, pipe)) {struct dbri_pipe *pptr = &dbri->pipes[pipe];
@@ -2506,18 +2517,6 @@ } } } - -static void dbri_debug_write (snd_info_entry_t * entry, -snd_info_buffer_t * buffer) -{ -char line[80 ]; -int i; - -if (snd_info_get_line (buffer, line,80 ) ==0 ) { -sscanf (line,"%d\n" , &i); - dbri_debug = i &0x3f ; - } -}#endif void snd_dbri_proc (snd_dbri_t * dbri)
@@ -2531,9 +2530,7 @@#ifdef DBRI_DEBUG err =snd_card_proc_new (dbri->card,"debug" , &entry);snd_info_set_text_ops (entry, dbri,4096 , dbri_debug_read); - entry->mode =S_IFREG |S_IRUGO |S_IWUSR ;/* Writable for root */ - entry->c.text.write_size =256 ; - entry->c.text.write = dbri_debug_write; + entry->mode =S_IFREG |S_IRUGO ;/* Readable only. */ #endif }
@@ -2637,7 +2634,11 @@return -ENOENT ; } -prom_getproperty (prom_node,"intr" , (char *)&irq,sizeof (irq)); + err =prom_getproperty (prom_node,"intr" , (char *)&irq,sizeof (irq)); +if (err <0 ) { +printk (KERN_ERR "DBRI-%d: Firmware node lacks IRQ property.\n" , dev); +return -ENODEV ; + } card =snd_card_new (index[dev], id[dev],THIS_MODULE ,sizeof (snd_dbri_t));
-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
Takashi Iwai
AppliedAt Sat, 10 Sep 2005 13:45:11 +0100,
Quoted 23 lines · Martin Habets
This patch contains the following fixes to the Alsa DBRI driver:
- Remove 2.6.13 build warning on the
prom_getproperty()call.- Rework command synchronization: send a sequence number with
D_WAIT, and check it in the completion interrupt.
Move synchronization delays from_cmdsend()to_cmdlock()allowing the CPU to do other usefull things while the DBRI is processing the commands.- Fix first argument of
printk()calls.- Enable burst transfers for DBRI. Original 2.4 patch from Krzysztof Helt
- Make
dbri_debugmodule parameter writable from sysfs. Remove obsolete write access to the /proc debug file.- Replace
udelay()withmsleep_interruptible()where possible.- Update documentation comments.
Quoted 7 lines
This diff is against 2.6.13, i.e. ALSA 1.0.9b. I can also send each fix split out if you want. Please apply.
Thanks,
MartinSigned-off-by Martin Habets <[email protected]>
Thanks, applied to ALSA CVS now.
Takashi
-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf